Written by Megan Jenkins on May 21, 2019
February 7th, 2018
approx reading time
6 Minute Read
With the development of technology, FEM and CFD simulations have become increasingly reliable over the years. However, the biggest drawback—the computing times when the problems involve nonlinearities or transient phenomenon—has remained unsolved. Real-time FEM, especially for surgical applications, has shown great potential but remained an academic technology primarily due to the computational times. In this article, we discuss five important tips that will help you simulate faster using SimScale.
Many problems, when idealized, involve symmetry. This symmetry could be along an axis or could be about a line. Using symmetry is one of the very common and yet powerful ways of reducing the overall size of the problem, thus increasing the speed of the simulations. Symmetry is said to exist not only in the geometry but also in the loads and constraints about a line or a plane of symmetry.
As shown in Figure 1, symmetry is most often found in nature. The same plate with a hole could also be a matrix with a particle. In such a scenario, the hole can be replaced by the particle material. Depending on the type of problem, considering symmetry would mean that the simulation runs 3-4 times faster than otherwise.
The purpose of a helmet is to protect the person who wears it from a head injury in case of impact. In this project, the impact of a human skull with and without a helmet was simulated with a nonlinear dynamic analysis. Download this case study.
A good mesh can make life much easier and simulations much faster. There are several factors to consider when meshing. A mesh convergence analysis is very much necessary to ensure that the obtained results are accurate. One of the ways to ensure a quality mesh is through mesh refinement. SimScale offers a mesh refinement option during mesh. This allows selected regions to be refined in comparison to refining the entire model, thus ensuring faster simulations.
Similarly, in the presence of sharp points and corners, the mesh needs to be refined in these regions. These regions can be easily selected to ensure mesh refinement, as shown above.
Most often, the CAD models are obtained from online repositories like GrabCAD and others. We advise caution in using these models, especially for fluid mechanics simulations. The outer surfaces frequently have engravings of copyrights or names of the authors. Even these minor engravings can result in small surfaces that significantly affect the simulation and increase the computational times of the simulation. It is recommended that these models are checked for such small surfaces to ensure faster simulations.
A more detailed analysis of meshing in structural mechanics problems is found in this blog article: “How to Mesh your CAD Model for Structural Analysis (FEA)“.
Another important aspect to be considered in structural simulations is the stability of the material model or parameters considered. Material parameters that are not unconditionally stable is the reason why structural simulations fail in 90% of the cases. The material parameters are generally obtained by fitting the experimental data. The fitting is limited by the maximum stretch to which the data is available. If the model is used beyond this limit, it might not be unconditionally stable. In such a case, each step in the simulation can take longer to converge and hence is slower to compute.
For example, let’s say the experimental data was only available for a stretch to 30% and the model parameters were fitted with this. When a simulation is done using this model, where the strains are much larger (> 30%), it is possible that the nonlinear behavior beyond 30% is not accurately captured.
Secondly, depending on the number of tests used for fitting, it is possible that the material demonstrates an instability when subjected to a different type of loading. For example, the material parameters for the Mooney-Rivlin model are fitted using only uniaxial tests. When a biaxial simulation is complete, it is possible that the results show significant instabilities.
Such instabilities can be identified by plotting the force-displacement diagram on simple tests performed using the chosen material parameters. Identifying such instabilities through simple tests like uniaxial, biaxial, shear, volumetric tests could help create significantly faster simulations.
Today, there are no more problems related to linear FEM that cannot be solved. We can confidently say that any problem that does not have a nonlinearity, i.e. material nonlinearity (like hyperelasticity or plasticity) or geometric nonlinearity (like large rotations in thin structures) or nonlinearity in boundary condition (like contact) can definitely be solved without batting an eyelid.
The biggest challenge for the coming decade is in solving nonlinear problems. In particular, contact is known to be the worst of the kind and is known as a “strong nonlinearity”. This means that it is like a switch, it is either on or off. The original mathematics of FEM was based on smoothness, but when there is no intermediate state, this results in a jump. This is also one of the reasons that contact remains an enigma, even after 300 years of study. A more detailed discussion on contact can be found in the article “Contact Mechanics and Friction: Is CAE Shedding Light on the Problem?“.
Contact has several applications, as shown in Figure 4, and there are no structural mechanics problems that do not involve contact. A good simplification to ensure faster simulations is to substitute contact with simpler conditions. If two surfaces are always in contact, then bonded contact could be a good option. Alternatively, force or displacement boundary conditions are often used to replace contact conditions. Depending on the problem of choice, one of these methods will undoubtedly ensure faster convergence and enhanced simulation times.
Parallel computing has been under a veil of complexity for several years now. It was primarily an area of interest to computer engineers but has been rapidly utilized by simulation scientists. Yet, the term “parallel computing” itself is associated with terms like “message passing interface” among others, which makes the situation claustrophobic for many mechanical engineers.
SimScale offers a great option for those who want to use parallel computing without having to worry about the internal workings. As illustrated in the Fig. 05, the number of computing cores can be easily set through a simple drop-down menu.
Figure 05 demonstrates that for both CFD and FEM. SimScale offers parallel computing with up to 32 cores. The use of parallel computing is strongly recommended, especially for large problems, as a method to reduce the overall computing times.
There are several more avenues to explore when it comes to increasing speed. Most often, civil engineers use simple simulations to get an order of magnitude estimate. Later on in the process, more detailed simulations are done on a reduced sample set. SimScale offers the possibility of running engineering simulations in parallel, and this option enables you to get the best of what the platform has to offer.
SimScale’s CEO David Heiny tests the capabilities of the platform to solve a real-life engineering problem. Fill in the form and watch this free webinar to learn more!
Subscribe to the SimScale Blog
Written by Megan Jenkins on May 21, 2019
This computational wind engineering guide covers types of wind analysis, how to improve your designs, and more wind resources...